Curved surface nodal (CSN) method is a novel approach to solve the neutron transport equation with curved surface nodes. The method employs multidimensional polynomials in space to expand angular neutron flux, with expansion coefficients being solved by constructing a linear system using weighted residual equations and integral leakage conditions. However, present research indicates that CSN method may exhibit divergence. The divergence is attributed to the current implementation which only ensures the consistency of integral averaged angular neutron flux in nodes’ boundary, failing to accurately reflect the neutron angular flux continuity at boundaries. This paper proposes two schemes to enhance the boundary conditions, namely endpoint constrained optimization and strict constrained optimization. The effects of these two optimization schemes indicate that strengthening the constraints on boundary conditions can effectively improve CSN convergence. However, the addition of equations leads to an over-determined linear system, which cannot guarantee computational accuracy. Therefore, increasing the number of unknowns and appropriately supplementing the boundary condition constraints to avoid over-determination or under-determination would be a viable approach.